The Agent Skills Directory

[SAFE] (SAFE): All analyzed files are markdown documents containing technical documentation, dataset descriptions, and benign Python code snippets for using the TorchDrug library.
[DATA_EXPOSURE] (SAFE): Code examples include paths to local dataset directories (e.g., ~/kg-datasets/, ~/molecule-datasets/). These are standard placeholders for machine learning datasets and do not expose sensitive system credentials or configurations.
[INDIRECT_PROMPT_INJECTION] (SAFE): While the documentation describes how the agent would ingest molecular or protein data, the context is purely scientific and lacks any patterns suggesting exploitable capability tiers or missing sanitization markers.
Ingestion points: Dataset loading via torchdrug.datasets and PDB file loading in protein_modeling.md.
Boundary markers: Not applicable in reference documentation.
Capability inventory: Local file system read access for scientific data.
Sanitization: Handled by the underlying TorchDrug library parsing logic.
[REMOTE_CODE_EXECUTION] (SAFE): No remote execution patterns (such as curl | bash) or dynamic execution functions (eval, exec) were found. Mentions of pre-trained models like ESM-1b are standard references for protein sequence analysis.

torchdrug